inspired by LAVOIR Β· arXiv:2609.30706

The Ask-or-Bet Line

A live-betting decision toy for the value-of-information tradeoff. Every slider tick recomputes when it's actually worth pausing to gather more information versus locking in your bet right now β€” the same amortized calculation LAVOIR trains a model to spit out in a single forward pass, instead of re-deriving it every time.

VoI(t) = 𝔼[value after signal] βˆ’ 𝔼[value now] βˆ’ cost(waiting until t) Paper: "LAVOIR: Teaching a Single-Pass Decision Encoder When and What to Ask" β†—

Situation

β€”
Your perceived edge over the current line, before any new information. Negative = you wouldn't bet at all right now.
β€”
How wobbly that edge estimate is (std. dev., in edge points). Higher = a signal has more room to actually change your mind.
β€”
How much of that uncertainty the incoming signal (injury news, lineup card, weather call) resolves if you wait for it.
β€”
Expected CLV erosion per minute β€” the line moving against you, liquidity drying up, or juice worsening while you sit on your hands.
β€”
Your hard deadline β€” tip-off, kickoff, or whenever the book stops taking this number.

Net value of information vs. wait time

waiting ahead bet-now ahead optimal ask moment
Worth waiting from
β€”
first crossover, decay start to bite later
Optimal ask moment
β€”
peak net VoI in the window
Stop waiting by
β€”
second crossover β€” bet now beats it after this

The race: bet now vs. wait for the signal

Bet now
β€”
Locks in your current edge immediately. No decay, no new information β€” what you see is what you get.
Wait for signal
β€”
Expected edge after the signal resolves, net of the CLV you bled while waiting for it.
t = 0WAITt = T
β€”